In this episode of The Next Next, host Jason Jacobs interviews Anis Bennaceur, co-founder and CEO of Attention.com, a company that automates sales operations using AI agents. They discuss the challenges of Anis' first startup, the breakthrough with GPT-3 that led to the creation of Attention.com, and how the platform addresses CRM filling pains. Anis shares insights on the evolving role of AI agents, the importance of quickly launching products, and balancing innovation with practical application. They also delve into using AI internally for building and marketing products, the transformation of CRM systems, and the future of AI in business and software. Jason and Anis explore the conceptual and practical aspects of using AI agents, customer engagement strategies, and the efficient use of AI tools for product development and market outreach.
Transforming CRM with AI: Anis Bennaceur of Attention.com
In this episode of 'The Next Next,' Jason Jacobs interviews Anis Bennaceur, co-founder and CEO of Attention.com, a company that automates sales operations through AI agents. They discuss the challenges of Anis’ first startup and the breakthrough moment with GPT-3 that led to the creation of Attention. The conversation covers how attention addresses CRM filling pains, the evolving role of AI agents, the balance between launching products quickly and maintaining innovation, and the future of AI in business software. Anis explains the practical applications of AI, internal uses at Attention.com, and offers insights on building AI-enabled workflows and autonomous agents. The episode also touches on company efficiency, hiring, future priorities, and the growing importance of AI in the tech industry.
00:00 Introduction to The Next Next
01:43 Meet Anis Bennaceur: CEO of Attention
02:51 The Genesis of Attention
03:56 Early Challenges and Breakthroughs
06:01 Pivoting to CRM Automation
09:54 Building and Launching the MVP
21:05 Defining and Utilizing AI Agents
28:04 Understanding Customer Needs and AI Autonomy
28:47 Challenges in Tool Selection and Flexibility
29:49 Building Features and Tooling Perspective
31:43 Relationships with AI Providers
32:53 Network Effects in B2B
36:08 Implications of AI on Product Strategy
39:09 Leveraging AI Internally
41:39 Autonomous AI Agents and Workflow Automation
48:02 Efficient Business Operations with AI
52:49 Key Priorities and Future Outlook
54:06 Final Thoughts and Shoutouts
Jason Jacobs: Today on The Next Next, our guest is Anis Bennaceur, co-founder and CEO of attention.com. attention.com automates sales operations using AI agents. In this episode we talk about the challenges of Anis' first startup, the breakthrough moment he had with GPT-3 that led to the idea for attention. We talk about how attention solves CRM filling pains, and anis also elaborates on the evolving role of AI agents, the importance of launching products quickly and balancing innovation with practical application.
And he also touches on. In the internal use of ai, both in terms of how they build out the product, but also how they go to market, the potential transformation of CRM systems, directionally, and the future of AI in business and in software. It's a great one, and I hope you enjoy it. But before we get [00:01:00] started.
I'm Jason Jacobs, and this is The Next Next. It's not really a show, it's more of a learning journey to explore how founders can build ambitious companies while being present for family and not compromising flexibility and control, and also how emerging AI tools can assist with that. Each week we bring on guests who are at the tip of the spear on redefining how ambitious companies get built.
And selfishly, the goal is for this to help me better understand how to do that myself while bringing all of you along for the ride. Not sure where this is gonna go, but it's gonna be fun.
Okay. Anis Bennaceur, welcome to the show.
Anis Bennaceur: Thank you, Jason. How's it going?
Jason Jacobs: It's going well. How you doing?
Anis Bennaceur: I'm doing great. Sunny day in New York. Can't be happier. Things are going good.
Jason Jacobs: It's also a sunny day in Boston. I feel like the weather in Boston and New [00:02:00] York, they don't always go together, but they tend, they frequently tend to go together it seems. But yeah. I'm psyched that you're making the time to do this. I got to you because when I published with Hadley Harris from eac I asked him if he had any portfolio founders he was excited about in these areas that might want to come on, and he pointed me to you.
And we have not done a prep call. I did do some prep heading into this discussion, but other than the five minutes that we just chatted before hitting record, it's the first time we're talking and we'll do it live as they say.
Anis Bennaceur: Thank you so much for having me. I had Lisa Best, by the way. Shout out to him.
Jason Jacobs: He's great. Yeah. Yeah. As we were chatting I, I didn't know him before, although I know some of his partners and a bunch of his portfolio founders. But it turns out, yeah, he grew up in this area. He went to one of the schools that, that my son's probably gonna look at for high school. And yeah, just a bunch of commonality.
And he see's a really great guy. But at any rate, tell me about attention. What is it, how did where'd the idea come from? How'd you get started in 10 words or less?[00:03:00]
Anis Bennaceur: Yeah, a hundred percent. Attention automates work out of your sales conversations in one sentence, right? So you can see as your system of ai.
Jason Jacobs: really mean one sentence. Take a, actually take as many sentences as you want. Yeah.
Anis Bennaceur: Yeah. So attention is your system of AI agents that captures all of your customer interactions. Think of your calls, emails, any, structured or structured types of interactions with prospects. Then all of these interactions are fueled into an insights engine, right? How do we tell you how your deals are going?
How is your, what does your pipeline look like? What are your customers saying? And and then the last piece is the agent piece, which is how do we push these intelligent insights to the right person at the right place, at the right time? And so this is what attention does, right? We automate your, all of your sales operations from the first meeting that you have until you close and you expand the client.
How do we get into that? Great question. So this is [00:04:00] my second startup. I'm a second time founder, and the last startup that I founded called Mixer, which is a professional network for creatives. We were actually selling subscriptions to enterprise firms like Capital One square, Havas Media and so on.
And. I was exposed directly to all the challenges that we're solving today. And the first thing that I remember very well is that, we bought a Salesforce subscription and Salesforce was always empty, right? And then in 2020, the pandemic happens and everyone starts working from home.
And it was extremely hard to know what was happening in our deals. There was literally no visibility. And so at that point I started thinking about these things quite a bit. There were already existing platforms back then that could record your calls, but nothing that could really automate any of this.
And really the big unlock was in July July 19th, 2020. GBT three comes out in, in the they had this playground [00:05:00] on open AI's platform. And I probably start playing with it and I could immediately see, some some pretty impactful stuff. And I think this is also when Jasper not too long later took off.
And b LMS were already starting to be pretty useful. And I started realizing that you could feed a call transcript into the lms. I saw that there was a massive need. And as I started attention in 2021, September, 2021 with Mathias, who happened to be my biggest competitor at the time when I was running Mixer we saw a massive need about around auto filling the CRM automatically out of your customer calls.
Jason Jacobs: Got it. And once you identified that area as an interesting one. What were those earliest innings like what did you do? Where did you start? How did you start navigating the inevitable chicken and egg? That happens whenever you're trying to birth a new thing [00:06:00] from zero.
Anis Bennaceur: Yeah, so I remember it. We had kinda like these ideas around the all the friction points for sales teams, but still, our past startups never got that big. And so it was important for us to jump into a lot more discovery conversations with with our ideal types of customers, right?
So we reach out to a ton of rev ops leaders and sales and revenue leaders. And overall the one thing that like came consistently around their biggest pains and priorities was filling the CRM, right? Initially what we wanted to solve was. Slightly different, which was more related to coaching, real time coaching for reps.
And we ended up realizing relatively fast that this was going to be more of a nice to have rather than a s had. But also at the same time, most it was going to be extremely difficult to release a a product that could [00:07:00] perfectly tell you what you had to say to the prospects in real time, especially given, the the state of tech back in 2021.
And so one of the other things that we kept hearing from prospects was we would pay you a lot of money if you could fill med back in our Salesforce if you could custom fill every single CRM field in our Salesforce. And at the same time, you look at, back in the day platforms like Julian and Scratchpad, they built relatively sizable businesses just surround the idea that you.
Could you could just fill the CRM with a nicer interface without even using ai. And so that was also a massive signal that, filling the CRM was a huge pain for every single team and including our sales in our last in, in our past lives. And so the second we started feeding the transcripts into GBT and then auto filling the CRM with that we saw the very early beginnings [00:08:00] of a product market fit.
Jason Jacobs: Got it. And when you think about those earliest innings how did you, determine tools to leverage off the shelf versus DIY? And then same question in terms of if you were to get going today in 2025, what would be the same and what would be different if you were starting now?
Anis Bennaceur: Yeah, a hundred percent. I think it was not as obvious at the time that GT three was going to be the first the first L that we were gonna use. And we we spent a good amount of time looking into the open source models that were on hugging face at the time. And so we tried a bunch of different things and the reality was just that GT three gave us the best possible outputs.
That was the first thing. And then the second thing was, just overnight. GP three had this submodel called DaVinci two, and then the minute it got to DaVinci three, the [00:09:00] quality of the outcomes was much, much better. And so we didn't really have to change much in our product. We just had to change a few lines of code.
And that allowed us to give much higher quality at outputs for our users. And at that point, they I remember very well the face of a user when he hit Analyze Call and he gave him incredible at outputs at the time, right? And he was like, this is wizardry, right? And and at that point we started, doubling down on open AI's Tech as, the main LLM and today, actually, as a matter of fact, we also use anthropic and and open ais. And we just let our users decide themselves which model they wanna lock in, or we optimize it based on the use case. Sometimes if they don't they're more agnostic.
Jason Jacobs: And how did you balance getting a product out the door or initial MVP [00:10:00] with making sure that you had coverage across categories that people cared about and that it was of a quality level that wouldn't be embarrassing. And I guess that's also just a founder philosophy question. How do you think about what the right balance is when you're bringing a new product to market?
Anis Bennaceur: Yeah, I think we really subscribed to the YC ethos over here, even though we didn't go through ic. I think a lot of it is very valuable advice. And so just launch now, right? Launch as fast as you can. Don't wait too long. Don't try to build anything perfect at once because you'll, it, it is better to launch fast and get users and then have some early tech debt and rewrite it as, rewrite your code as you grow.
But for us it was basically talk to as many users as you can and write code, right? And so the more you talk to users, then you try to understand their biggest pains, how to even design the [00:11:00] product. You come up with mockups, right? And then you get back and I call them and show them these mockups.
And if it validates anything, then at that point you can start building in that direction. One thing that we also a great growth a great product hack actually, that my co-founder came up with at the time, credit to him was. Try to find a product that exists that solves what people are telling you, right?
And once you find that product try to sell it to them. And so we found we had this one point, this idea of capturing capturing so through facial recognition, capturing people's emotions. And there was one product that existed that did that. And I signed up to the product and then tried to sell it and I realized that people did not wanna buy it.
That really expedited six months of going the wrong direction. And so I think that's a great hack, right? You find a very early stage product [00:12:00] that other people have built. You say that it is your product, you try to sell it and see how people are reacting to it. And
Jason Jacobs: heard that before. I like that a lot.
Anis Bennaceur: yeah. But, so yeah, that really saved us in terms of not having to go in the wrong direction.
And then over time, you start getting some users very quickly. You have to be very honest to yourself. And if you're seeing that they're not, they don't love your product, they don't get back to you very quickly. They don't scream at you for something that could be improved, then you probably built something that people don't care about enough.
And very fast, when we had our very first product out, that was real time coaching. Our first design partner, their reps did not even care about the product. We tried to get them on calls to give them, give us some feedback. You know what? Half of them did not even show up to be feedback calls.
The other half were like, Hey, look, this is not that great. The feedback is not [00:13:00] very helpful. We thought to ourselves are we gonna spend the next few months trying to improve this and get this to product market fit? Or do we iterate towards something either more adjacent in the same space that could be more useful, or do we just complete pivot?
And the other piece was when we went back to the drawing board, we started hearing more and more around the pains of filling the CRM. And so within, I think it was about four weeks, we decided that we were gonna build something that filled the CRM instead and completely stopped innovating on the real time coaching piece.
And that was the best decision, right? We start seeing it. It was like this trifecta of no one was doing it. It was a massive pain. Other company, it was a step function in terms of improvements compared to what, legacy tools were doing. Like Scratch, bad and Duly, where you had to manually input the data.
[00:14:00] Whereas here it was fully automated, right? And so that wedge, we were the first company to autofill the CRM basis based on conversations. And that lasted long enough to give us a bit of an advance to then build more raise more money, and now build, go from a point solution to a compound product.
Jason Jacobs: How did you think about what customer you were building for in those earliest days? So as you were turning the corner and focusing on the CRM what was it? Was it based on size? Was it based on industry? Was it based on geography? How did you figure out what it was based on? And how specific did you feel like it needed to be before you put the product into the world?
Anis Bennaceur: Yeah I'd actually tell any founder not to overthink it right back. Back then, we were just trying to sell to, the only thing that we knew is that we wanted to sell to B2B sales teams. And we didn't care about the size as much. And as a matter of fact, our [00:15:00] first design partners were tiny companies, right?
CT series A startups actually never more than, never bigger than Series A startups, because no one else will actually give you a chance, right? If you go to a company like Ramp or Brex, or. And, they have such well-defined processes, you don't want to mess up with 'em. And I remember when we actually launched our product to the market within week three, we had one of the largest US companies that came inbound and wanted to buy our solution.
And then we received a 25 sheet InfoSec Excel doc that we had to fill. And, over time their as we filled it and sent it back, their InfoSec got back to us and said that they, we could not, we should, we could not be connecting to their calls and emails and so on and their calendars and blocked us.
And in hindsight, that was the best thing that could happen to us because we were just not [00:16:00] equipped to be serving such a massive enterprise firm. Now we definitely can, but back in the day we couldn't. I would advise any very early stage startup to just try to build for the smallest possible customers at first, and then graduate over time to larger and larger clients.
When you think about our ACVs, I think from the moment we launched until now, our average a CV was multiplied by 10, in two years. And so the whole point behind this is that try to build for users that will be forgiving if you mess things up and learn with 'em. And then over time try to close higher and higher a CV types of clients.
And at that point you'll start thinking about maybe more international teams and more complex types of businesses. But do not try to go for enterprise right off the bat, for example.
Jason Jacobs: And I, thing that's fuzzy [00:17:00] for me is I'm still sorting through. It's okay, so if you use AI, for example, it's like you can use an LLM or you can use your own model. And then it's if you use an agent, it's oh, now the agent's actually out doing it. So it's not you using an LLM, it's the agent that's out doing stuff.
I've heard you talk about on other shows that you have a fleet of agents in house, and I guess my question is when it came to building the product that's going to listen to these calls and then populate the CRM, it seems like there's a chicken egg because you need the data, but where does the data come from?
And then once you have the data, is it you that does it for a period of time and then you're training a model to do it? And then where do agents fit in? I guess just walk me through practically, I guess one is just definitionally how you think about what the stuff means, and then two which aspects you do, which aspects
Anis Bennaceur: Yeah.
Jason Jacobs: machines do, and then how that pendulum shifts over time.
If at all.
Anis Bennaceur: Yeah. So [00:18:00] as a business at the application later, we don't train models. And I think a year or two ago, a lot of VCs were getting this wrong, where they were telling you, Hey you're just a GBT wrapper or an open AI wrapper. The reality is that as a business at the application layer, you should not train a single model.
You should spend time on improving the infrastructure of your product, on improving the ui, on making sure that you're building things that users want. And so some of the areas where we are spending time on is is more around the the retrieval augmented generation around.
So how you provide context around certain prompts, right? Or within your product. So let's say you have a product for sales for a specific type of business, you may want to describe everything about the business and how that sales team [00:19:00] operates so that, certain agents will give you the output that you're looking for.
So that's like the world where most startups at the application layer are at today. And then over time, yes, when you're having, agent to agent interoperability, they will speak to each other. There's a lot of. Thinking around how you take the unstructured data and restructure it into specific specific outputs, whether it's through A-J-S-O-N or whether it's however you wanna structure it.
And as you restructure that data, that's what could get fed from one place to another? The whole point here is that you won't, you shouldn't really think too much about how you're going to train any models or these agents should be talking to one another. I think that there's a lot being built at the infra level around, around the models getting better.
And now, [00:20:00] with MCP and also A two A, but the protocols that are getting adopted, those are also going to get better and better. And so you may not want to go against the wave there, you. Wanna ride the wave the right way. And so you want to think about how these different models and protocols are going to get to, to keep getting significantly better over the years, and how you should build a platform that only keeps getting better as these models get better.
If there's a new model that comes out, you should be able to swap out the old model with a new model in certain cases. Or allow your users to use the latest model as well if they want to.
Jason Jacobs: And then where do agents fit into all of this? It'd be great to understand just your
Anis Bennaceur: Yep.
Jason Jacobs: journey. Like how did you start with agents? What was the use case? did you figure out whether to build them in-house or use the off the shelf agent building tools? And and what have you [00:21:00] learned along the way that might be useful for others that are not as far down the path?
Anis Bennaceur: Yeah. I might be wrong, but the way we're thinking about it here is that so first of all, let's define an agent, right? Because I think that's something that is fuzzy to everyone. What is the difference between a workflow or an AI enabled workflow and an agent?
And I do wanna be clear about this. An AI enabled workflow, you'll tell, Hey, these are all the different steps that I want you to follow. And you may use AI in some of the steps, but it's very deterministic. An agent actually. You have what we call here or at least within our internal frameworks, you have three different layer levels of agents, the first ones, and over time they'll get more and more autonomous.
But I think the world is right now to a point where you can tell an agent, I want you to find this for me, right? Or I want you to do this for me. And the agents will be able to define the [00:22:00] steps themselves without you telling them what those steps are, and it will most likely get it right. So it will define every single steps without you building every single edge, case and loop and if else statements and so on.
I think the world is gonna get to a point where you, so a le a level above where the agents, you actually just give them a problem and they go out and solve it for you. So something like, Hey, our conversion rates have dropped 10%, go figure out why and do it for me. And so at that point, the agents will be able to build all these systems and processes to give you what you're looking for and solve that problem for you.
And then the last piece, which is like the holy grail is an agent because it's all connected to every single system that you have. Being able to identify the right problem, tell you, Hey, I actually noticed that 10% that, that your conversion rates have dropped by 10% [00:23:00] and these are all the systems and processes that I built for you, and I think this is where the world's gonna get to.
Now that we define what an agent is, I think how do you make sure that you build the right agents? I think today a lot of companies are rushing towards building agent builders that allow you to build the first layer that I just spoke about, right? Which is, I want you to build this, figure out how to build it.
The second piece, I think as we get to the second and third levels, a lot of companies will have figured out how to have one agent builder speak to another agent builder, right? Let's say the attention agent builder speak to the clay agent, lip builder to the Zapier agent builder, and so on. But I do think one thing that is very important is how do you think about your specific, or not just agent builder speaking to agent builders, [00:24:00] but what I actually kinda meant is attention agents speaking to clay agents, speaking to Zapier agents, speaking to all sorts of agents.
And so there will be more and more adopted protocols around how to do the, do this well, I think just the internet was created. I think you'll, I know that Google released a two A as a protocol. I don't know much about it, but I know that this will, this could help with this. So the whole point here is that how do you adopt the best agent builder for your use case?
In our case, we're building something for the bottom of funnel and attention might be the best possible platform today to build AI agents for your bottom of funnel sales. Maybe Clay could be your best, AI agent builder in a few years, or even a year from now around your top of funnel, right?
Prospection or Unify, right? Which is also another great company at the top of the [00:25:00] funnel. They could be your best company to build AI agents that capture your signals and then turn those into outbound. And so we could have, we could see a future where a user has their attention, AI agents, their unified AI agents, and they all speak to each other and they would be speaking to each other via these protocols.
Does that I feel like I I went pretty deep into that one, but does that make sense for you?
Jason Jacobs: It does. I think one of the one of the there's a lot of talk about how it's no code tools and you can just go as a non-engineer and you can but like understanding how these pieces fit together and then catching when there are hallucinations or mistakes or whatever. I, it, it, it seems clear that there needs to be that trained eye that oversees and I think what's also murky to me is for example, you might have said when you were [00:26:00] heading out, we're not just gonna build a a tool that populates the CRM, but we're gonna use this as the. Tip of the spear to build a new CRM, right? It or you or the company that you're using for to, for the voice to text and the context. They might say, Hey, now we're gonna add a user interface layer, right? That's gonna look more like what you're trying to do. That will actually interface with the voice to text and populate the CRM or the CMA CRM players might say, Hey, we're gonna move in this direction.
So I guess especially given how quickly the landscape is changing under all of those players feet, how do you think about where to focus, where to innovate, how to and how to just tap dance through the minefield that, that feels like it is switching dynamics with every passing day and week.
Anis Bennaceur: Yeah, good question. There's a ton of convergence, [00:27:00] right? Like you're seeing a lot of players that we would've never even thought could be competitors, could be competing with us on one feature or one use case that we have, for example. And so that's something that could happen. The reality is. I think if you hold as your Comcast just what customers want as the North Star I think you're going to be in good shape.
You could, I think a lot about what if, what if One Day OpenAI built this mega platform that just wipes out trillions of dollars of software VA value through building something that just aggregates everything, aggregates all the use cases into one platform, right? You're starting to see some, a few steps in that direction.
Right now you can connect your Google drive to chat GBT, and you can ask questions across your files, which is like the GLE use case. And you're seeing, so in other words, you're seeing [00:28:00] infra players playing more of the application layer, which is really interesting. But the initial point around the direction to take, I think that if you keep in mind what customers want, and not necessarily just the features that they ask you to build, but actually the pains that they're dealing with, and it is your responsibility as a founder to reinvent from a UI perspective and UX perspective, what that's going to look like.
Then you're going to be going doing great things. Now, an overall direction that we're taking is just more and more autonomy within the AI agents that I just described earlier, right? So going from like level one to level three of agency and what we're building, I think we're gonna be in a great shape.
Jason Jacobs: Uhhuh. And and I guess similar question around the tool selection under your feet, right? Because I've heard from anecdotally that can be hard where sometimes if you build a [00:29:00] feature based on an existing tool set, but at the time you launch the feature, the tool set under your feet has changed and the, and your architecture decisions, for example, be outdated.
How do you. Keep up in, in, how do you make any decision? Is there a such thing as for the long term anymore? And I guess what I mean by that is someone I bought my car, right? And and someone said gosh, that was dumb because the technology's changing so fast. Like you're better off with the lease because by the time that lease is up, your car's gonna be so outdated relative to what's available at that time.
Like, why would you wanna hold a car for the long term? And so is it, is that also happening with everything and does that manifest with with all the infrastructure and vendors and and tool sets that you're that you're using to, as the foundation for your offering?
Anis Bennaceur: Yeah, that's a great question. That's, it's really hard, right? It's extremely hard and painful and we think at some point we should build this feature this way. And then MCP comes out and [00:30:00] makes all the tech that you've built for a year or two complete obsolete, and then maybe there to be something else that MCP, coming out in a year from now that will make everything that you build now complete obsolete.
And the reality is I think teams will have to be more and more flexible around and being comfortable around ripping and replacing versus having this kind of big ego around the proprietary tech that they built, right? You might and I think that's where our team, our engineering team has done a great job here at like always questioning the foundations that we built software on, go back to first principles and if there's something that will make 10 x different in the features that we're serving our clients, then we should probably go in and rewrite those features, right?
So that's from a feature building perspective. From a tooling perspective, look as independently from our product. I [00:31:00] think from a, let's say from a go to market side of things, we're always reevaluating our tools and always looking out for new vendors and always looking at what's best in the market.
So if there's something better, we'll go out there and ripping, replace or the. The tooling.
Jason Jacobs: Back when I, I, as we discussed before we hit record, I happened to build a company that was, built a mobile app that was in the app store right at launch. And then when Android came out, we were there for the Android launch. And so we ended up spending a bunch of time ongoing interfacing with, with the Android developer relations and with the Apple. Developer relations, does that dynamic exist at all with the LLMs?
Anis Bennaceur: Yeah. So we have relationships actually with Anthropic and open ai, for example. So that definitely exists. In our case, the way that we've been doing this is we started out with one element that we're using, right? With open ais and then we included [00:32:00] Philanthropics. And we're doing the same, by the way, with transcription providers too.
And the whole point is that, a, you may want to build. Or the way we built things is we have an infrastructure that either optimizes for ELM or the, let's say transcription provider, for example based on the use case. But also we have like this waterfall infra that will switch models if one is not available at a given point in time.
And so you may want to build something that is extremely flexible that could use different types of infrastructure providers based on the use case. And that allows you to swap to swap the LM again based on the new LMS that come out. And that so far has been working great for us.
Jason Jacobs: How do you think about network effects, if at all? Who do I.
Anis Bennaceur: Yeah. In the B2B world, it's [00:33:00] relatively difficult, right? So depends on how we define network effects. If it's from a people perspective, it is relatively hard just because of privacy in the B2B world. However, from a data perspective I'm not probably the best one at talking about this, but it's all about how you define your evals implicit evals, right?
The way people build things, the way people, build their workflows or they a, their agents and how they define success and outcomes. I think this is going to be the next type of network effect, which is I. And by the way, we see a box. Aaron Levi did a great post about this yesterday, which is like how we're thinking about it.
You have your inputs, then you have your, the insights that are getting generated, and then you have the AI agents who are taking actions on your behalf. And then you have to record the outcomes, right? And the [00:34:00] outcomes and I'll give you an example in a second, but then once you have all of these that are getting tracked, then you start having a flywheel of success definition over time through propensity models or however you wanna think about it.
That keep then when every time you're successful at something, keep tightening that feedback loop with the types of insights that are getting generated and then the types of actions that should be taken, right? And so that's like how you get a flywheel of success. So I'll give you an, a very simple example.
Let's say you have an insight out, out of a deal that says, this is how you should build a a customer business case right after, after the third or the fourth call, right? An AI agent will build that for you. You'll share it with your clients, and then over time you'll start capturing success, right?
Success being, [00:35:00] did this convert to a call, another call in the process, or has this deal been marked as close one in your CRM? So over time you get so much da, so much data that you can note at a specific point in time that if you share this type of business case at call number four, instead of call number two, that might make you more successful.
And so you may your system will start making smarter and smarter recommendations and acting in a compounding intelligent way for your users to be able to make them more successful. And so you end up closing like that, that, that gap, that knowledge gap for teams, and you're starting to make that, you'll start to make them more and more efficient over time.
Jason Jacobs: Huh? Yeah. Some of the stuff I've been thinking about is around youth sports skills training, and already the light bulbs are going off in my head as you're talking, it's oh, wait, if you watch more and more kids train, you can actually start to. [00:36:00] Predict what types of training for what type of kid will yield the biggest types of returns for that skillset.
Which is really cool. I guess a semi-related question, we talked before about competition and how it's getting easier and easier to build and there's all this convergence happening. What are the implications of that as it relates to narrow and focused versus your footprint, tackling new hills, evolving from app to platform?
Like all the kind of the the VCs of of a decade ago, right? Are they outdated now? Be because of ai? Like how is it changing product strategy?
Anis Bennaceur: Yeah, that's a great question. That's something that we think a lot about. So there's two things to, or one main thing to keep in mind here is that it's been easier and easier to build software than ever, and it's going to keep going in that direction. And so the way you want to think about it [00:37:00] is how deep can you go into one specific use case or feature to the point where.
There's very little improvements left to be done, right? Maybe if you get a competitor that comes in within that specific use case, at best they could be 10% better than you, but that won't matter enough to get a customer to churn you to go to that competitor. And once you're done with that use case, then you move to the most adjacent underserved use case and start to build it in a similar fashion, right?
And you don't want to go into anything that is too far beyond, right? And I'll give you an example. Let's say in our case, we are very good at automating the bottom funnel, right? Filling the CRM, scoring your sales reps scoring your pipeline, and so on. It won't be too crazy [00:38:00] of a jump for us to go from this to now automating the top of the funnel because.
A, it is too, not only it is too far for from our current use case, but also B, there are way too many companies that are incredibly good at doing that. So we don't even want to waste our time on, on going in that direction. And so the actual answer to me is building a compound startup within the most narrow scope that you can have.
Right? And that might not be that trivia or that easy to understand for people, but I'm sure in each use case you'll think about what is the best scope, the most narrow scope, where I can be incredibly good at. And that has to be balanced again, around what your customers want. Some customers will say, Hey, I want the best in class tool for top of funnel.
I want the best in class tool for bottom funnel. And then some other customers will say, I don't really care. I just want something that automates [00:39:00] everything from top to bottom. And I think in both cases, the markets are huge and you just want to think about what's best for your clients.
Jason Jacobs: I've been dying to ask you given how much AI plays a role in your product itself, are you leveraging AI internally as it relates to how you build? And then same question as it relates to how you go to market.
Anis Bennaceur: Yeah. Everyone at the company uses AI every single day, right? Like a hundred percent of our company. And every function. So our engineers obviously use ai every uses cursor. Then everyone within growth and ops is also using AI to automate more and more growth types of outbound or content creation.
Every single day we use ai and then even our sales reps, right? They use attention. As an AI platform, but also sometimes like when, like GB oh three is [00:40:00] incredibly good at this point to even help you think about how to go a certain way, right? Ask it, asking it for advice. Like it's it's surprisingly good now compared to all the previous models.
And every single one, our company uses AI just every single day, multiple times a day, whether it is through chat, GBT or whether it's through specific solutions that are leveraging ai. And the last piece is how do you automate workflows or agents within specific use cases, right? I want to outbound to specific category of feeble, or I want to create content out of my sales conversations that, anonymized content out of specific use cases that our customers have and push it as block content, right? GT three will be incredibly good at doing that. And if you use a tool like Air Ops incredible platform for content creation, you'll have the ability [00:41:00] to publish tens, if not hundreds of articles that, used to take you days or weeks to write back in the day.
Jason Jacobs: That all makes sense to me as it relates to using AI in a point way. So the way I do it, like I record an audio of this discussion and the video, I will populate it in the script. I will use AI being descrip. To help edit it, I will River, we're using Riverside to record. I Riverside will pull the clips using AI of the ones it thinks is interesting.
It's like you click a button, you get an outcome. But earlier you were talking about how we're moving more and more like those layers of agents, right? And how we're moving to a world where agents will be operating more and more autonomously with with guidance that's like higher and higher order, right?
Or more and more self-directed, I should say. How does that get managed? Where does that sit in your organization? Where
Anis Bennaceur: Yeah.
Jason Jacobs: sit in any organization that starts to lean in this direction?
Anis Bennaceur: Yeah. So it is not [00:42:00] trivial, but I'll try to be creative here, in your specific use case. Something. So level one would be saying, Hey, every time I have a call or I have a podcast getting recorded with a guest, I want you to search everything about them, whether it's, so go out and transcribe their past podcast, read every blog post or LinkedIn post that they wrote about, and just give me everything that is out there so that I can come up with very smart questions for them.
So that's like the first level. The second level would be something around
you say, I'm gonna meet with this person. And then it figures out the entire system and process. It figures out the questions for you. It emails them the questions so that they can be ready and for the podcast. And then it could do some backs and forths before the podcast getting recorded.
The last level of autonomy if I were to think about it, it is the AI agent [00:43:00] saying to you, I've listened to all of your past podcasts and it does seem that if I were to send these specific emails to the guests before they come on your podcast and I speak, I interact back and forth with them that will give the best possible episode for your guests.
Is that crazy?
Jason Jacobs: Yeah. I think about, I sent, as we were talking before we hit record, I send these weekly updates on my journey as I'm trying to figure out what to build next, and then I'm kicking out two podcast episodes a week. been several months. So they're starting to add up. And and it's imagine if a model like, and whatever that means, a model, right? But something, someone ingested all of the, like all these long form discussions and then all those newsletters and help me figure it out, what I should do, right? It's like in some ways it might know me better than I know myself.
So
Anis Bennaceur: A hundred percent.
Jason Jacobs: Yeah. [00:44:00] But where, so that direction's exciting to me. What's hard for me though, is actually translating that from a theoretical vision dream to a practical reality. What tools should you use? What are the steps? What skill sets do you need around the table? How much data is enough data?
Just that, that's where I haven't quite gotten to that granular level yet, which is why all I do is talk about it and get excited. And I haven't actually done anything yet, eh.
Anis Bennaceur: I think that's a good point. Just I just get building and that's okay if some things could be obsolete in six months or a year, because at that point you'll know so much more about your users and who you're building for, and you could just go back and rip and replace whatever you built.
Jason Jacobs: Yeah. But do you, is it so straightforward that you hear that and it's oh, you should be using X and here's what the steps should be? Or is it like, oh, every case is so different that it would require me to bring [00:45:00] in an army of of McKinsey consultants to do an analysis and that like how do you, how do, how does one figure that out?
Anis Bennaceur: I honestly don't overthink. It is hard to balance sure how you build things, but the reality is, can I just build something that my users want and that should just be the North Star. If I'm building something that my users want, they don't care about what models, or most of them won't care about what models or what, how your infra is being built.
Just build something that is truly useful for them and that's it. I I wouldn't really overthink it. That could contradict a lot of things that I said earlier, but, if just what YT says is launch now and I would just launch something as fast as possible,
Jason Jacobs: So
Is it one set of agents that's doing one specific set of activities or do you have almost like clumps of agents that are doing different things? And is it the same person or team that's managing all the agents? Or is it more done by, by, by functional area?
Anis Bennaceur: Yeah. Great [00:46:00] question for, in the case of attention, we have templates and libraries or libraries of templates that could be adopted that then the most important thing is that each client has their data structured in different ways and their CRMs and specific custom objects and their businesses will run in a specific way and they'll have x amount of teams and these amount of reps.
And so you the, at the end of the day when we onboard a client, we have a team of. What we call technical account managers that, or for deployed engineers that will deeply understand how your business works and operates and build custom agents based off of the initial template libraries that we have.
I think over time, we will get, we could get to a world where these agents understand your business for you and can actually get built without the help of the four deploy deployed engineers.
Jason Jacobs: If [00:47:00] you were a betting man looking forwards will the set of CRM kings and Queens in this era be the same set of CRM Kings and Queens in the next era? Or or would the landscape look different? I.
Anis Bennaceur: Yeah, I think the leading CRMs will be seen as data warehouses in the future. And, this is the entire bet that attention is making, right? We want to become the system of cognition that sits on top of your CRM, right? Your CRM ends up being like a data warehouse, where you have all the data is like your spine, but then the super intelligence comes from a product like attention.
Jason Jacobs: Huh. So you have no aspirations to build a CRM if you it's more like adding a layer of intelligence and then sucking more and more of the intelligence away from the CRM over time so that the the end customers are reliant on you and not necessarily their CRM of record.
Anis Bennaceur: I I won't say much more here. [00:48:00] I.
Jason Jacobs: What, so given how much you're leaning into ai what are the implications as it relates to hiring, as it relates to capital efficiency, as it relates to timeline to exit or exit size? Like
Anis Bennaceur: Yeah.
Jason Jacobs: you think about just practically how it changes the odds or the the the risk calculation when you're determining how to fund the business, for example, or how to
Anis Bennaceur: I love that question. Yeah, I love that question. I think over here at Attention, we're incredibly as a business, we're incredibly efficient. Our AR per employee is very high. And the way we're thinking about it is that we really try not to hire within a specific function if things can be automated with ai.
And so I know that a lot of leaders, like Toia at shop Shopify, or I also saw the CEO Duolingo write something very similar this week. And that's been how we've been thinking for the [00:49:00] past two, three years. And even when I was building the business, I wanted to build a business that could be incredibly efficient.
Now, businesses will still have to grow in head count because of competitive plays, right? Your competitors will hire more people. They will probably outgrow you. You can maybe play in a smart way by automating more and more, but still over time you'll need to serve your customers and your prospects and so hire more headcount.
For example, the main difference is that I think a RR per employee will be way higher than it was, and that also means that companies will most likely IPO with fewer employees than ever. And that will make an entire new generation of early employees being way richer than ever, right? Through equity [00:50:00] allocation.
Jason Jacobs: Do, does that also mean that these companies will raise less capital while they're private?
Anis Bennaceur: It really depends on what you're building, right? Interestingly enough, so I kinda realized this the other day where I looked up the, all the employees that we had over at Attention and we're a, late series A, almost series B startup at this point. And 65% of our employee base is actually engineering and product, right?
Normally it's, it tends to be closer to 50 50, especially for a product at the application LA layer that serves revenue teams. I just think that you could be as as efficient as you can, the team, you most likely need to have more head count. That will take overview, a lot of these automations that you're running, build more automations and cover more and more.
That's just like [00:51:00] how we're thinking about it. I'll give you, I'll get into a specific example. In our growth team, we, for example, have only three people in our growth team. One covers outbound, one covers content, and one covers business operations. And they have a ton of automations and they're running this team.
As if, historically there were 10 employees within that specific team. And now we'll add another, content hire, but I don't wanna have 10 content hires that will create content for us. I only want to have one person that will manage their entire army of content agents.
Jason Jacobs: I mean you alluded to this a little bit before, but where does the line of demarcation stop as it relates to like in your wildest dreams, what have you built if you're successful and and then what is outside of the domain of where you would go, where you know [00:52:00] that it would make sense to either just say no or to partner.
Anis Bennaceur: Yeah. If we were to get attention to this, to the system of recognition that I mentioned earlier, that is fully autonomous and agent and do that way faster than everyone else. Like as in being years ahead of our second closest competitor and be worth tens of billions of dollars as a business, that'll be incredible.
And so that means serving all of the best logos in the world, all of the best in class customers, and meet that one obvious platform for every customer.
Does that answer your question?
Jason Jacobs: it does. Yeah. Yeah. I'm cognizant we're pushing up on time final few questions. So key priorities over the next 12 or 18 months.
Anis Bennaceur: Yeah. Continuously grow. So continuously grow the business from a revenue perspective and making sure that our [00:53:00] customers still love us as we grow the business, right? So that means investing a lot in the product and making sure that, we keep super serving our clients as if it was day one, but with a way higher quality product.
Series B will probably happen by then. And yeah it is getting so hard to know what the world will look like in a year. But
Jason Jacobs: be a venture investor right now. Like it's it's just so hard to know. Like the every, it's just uncertainty at every layer. It seems like
Anis Bennaceur: a hundred percent Yeah, I know. At the end of the day, I think, like we're like closing the loop, but when I think of Hadley I think one of the main reasons why he invested in us and me and Mathia is he just saw something in us as founders. He definitely made that founder bet.
We were both repeat founders and we were approaching things the right way, and that's how. He made that bet versus [00:54:00] trying to look into every single loophole that our business model was gonna have.
Jason Jacobs: Anyone speaking to the listeners for a minute anyone from the listeners that you want showing up in your inbox? Are there, customers or key hires that you wanna shout out? It is just a, a promotional chance, if there's any anything you're trying to solve for that you wanna call out to the world.
Anis Bennaceur: Wait, would you mind repeating that? Any clients that
Jason Jacobs: If you have if there's, if you're trying to make a key hire, if there's a certain type of customer you want to hear from, it's just anyone listening that you want to hear from, this is your chance to give them a shout out.
Anis Bennaceur: a hundred percent. From a key hire, I would love to have the perfect content. Hi, hire. We. Work five days a week here in New York. So it'll have to be someone who lives in New York and is okay to grind, actually wants to grind. From me, customer perspective after seeing that Toby LinkedIn post and Twitter post, I would love to have Shopify as a customer.[00:55:00]
Jason Jacobs: Okay. I know I, Toby actually was an avid listener of my last podcast. I don't know if he listens to this one, but maybe he's listening. And anything I didn't ask anise that you wish I did, or any parting words for listeners?
Anis Bennaceur: I think we've covered a ton. I'll just repeat maybe a few shout outs to other platforms that I think will do great things. They're actually also customers of ours and we're very happy to have them unify. Incredible platform for your top of funnel and capturing signals and then automating outbound for you.
And Air Ops also incredible platform to create content.
Jason Jacobs: Great. Thanks so much for coming on. I learned a ton
Anis Bennaceur: Thank you.
Jason Jacobs: yeah, I'll eagerly be watching your progress and looking forward to keeping in touch. I'll be rooting for you.
Anis Bennaceur: Thank you. Bye.
Jason Jacobs: Thank you for tuning into The Next Next. If you enjoyed it, you can subscribe from your favorite podcast player in addition to the podcast. Which typically [00:56:00] publishes weekly. There's also a weekly newsletter on Substack at the next next.substack.com. That's essentially for weekly accountability of the ground I'm covering, areas I'm tackling next, and where I could use some help as well.
And it's a great area to foster discussion and dialogue around the topics that we cover on the show. Thanks for tuning in. See you next week.